metadata
license: mit
language:
- fi
size_categories:
- 1K<n<10K
I added some new instructions that use the same topics as the FIN-Bench but those are just the topics which I still generate using the same prompt I used for generating the rest of the instructions. (also added some other unrelated instructions and removed/replaced some)
Another change is format. This now uses ShareGPT formatting.
I'll make a new dataset after this using a different prompt to get it a bit more conversational.
From the previous dataset: This is a larger version of my previous dataset containing the same data but now with a lot more instructions generated using GPT-4-Turbo and GPT-4o. It may have issues that I've not noticed.
It contains parts of the Stanford Alpaca seed tasks translated to Finnish using DeepL/ChatGPT with manual edits to fix any issues. It also contains some ChatGPT generated instructions.